Seemingly innocent optimization related to xs_bind() broke TCP port
reuse by making non-reserved ephermal socket port to not be saved
in "struct sock_xprt (srcport)". In case of non-reserved port,
allocation happens as part of kernel_connect() inside of
xs_tcp_finish_connecting(). kernel_connect() returns EINPROGRESS
and the code skips stashing srcport in sock_xprt for reconnects.
This affects servers DRC in case of network partition where client's
RPC recovery would try reconnecting with a different port.
Reported-by: Alexey Kuznetsov <alexeyk@xxxxxxxxxx>
Reviewed-by: Jacob Strauss <jsstraus@xxxxxxxxxx>
Reviewed-by: Alakesh Haloi <alakeshh@xxxxxxxxxx>
Signed-off-by: Vallish Vaidyeshwara <vallish@xxxxxxxxxx>
Fixes: 0f7a622c ("rpc: xs_bind - do not bind when requesting a random ephemeral port")
---
net/sunrpc/xprtsock.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index c8902f1..5bf75b3 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -2393,9 +2393,11 @@ static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
switch (ret) {
case 0:
- xs_set_srcport(transport, sock);
/* fall through */
case -EINPROGRESS:
+ /* Allocated port saved for reconnect */
+ xs_set_srcport(transport, sock);
+
/* SYN_SENT! */
if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
--
2.7.3.AMZN
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
